Clio's Learned Hand Deal Puts Lawyers and Judges on One AI Vendor

Clio, which sells AI to litigators, now owns the AI that judges in LA County, Michigan and 10 states use to draft bench memos and orders. Its only customer pledge is continued support, so courts should get data segregation, model disclosure and exit rights in writing.

By Rajesh Beri·September 30, 2026·9 min read
Share:
A judge's empty wooden bench in a quiet courtroom with a laptop open on it beside a stack of case files and a printed draft order, the lawyers' counsel table visible in the foreground with a second laptop.

Illustration generated using AI

If your court runs Learned Hand, the company that drafts your bench memos is now owned by a company that sells AI to the lawyers appearing before you. The only written commitment you have is that it will "continue supporting" your deployment. Clio announced on September 30, 2026 that it had acquired Learned Hand, the AI workspace that judges and clerks use to summarize filings, research disputed issues and draft orders. Terms were not disclosed. The announcement says nothing about separating court data from Clio's law-firm products, nothing about which models will run underneath, and nothing about what happens to a court that wants to leave. Those are now renewal questions, and court IT should ask them before the pilot converts.

This is not a story about a practice-management company buying a startup. It is a story about one vendor sitting on both sides of the bench.

What Did Clio Actually Buy?

Clio bought a live judicial product with named, high-profile courts already on it. Learned Hand connects to court case management systems and helps judges and clerks "work through filings and disputed issues," conduct legal research, and prepare "bench memoranda and draft orders tailored to procedural requirements," per Clio's release. Founder Shlomo Klapper describes it as a "reasoning engine" for judges that organizes case materials, drafts orders — and flags "when lawyers bend the truth."

The customer list is the asset:

  • The Michigan Supreme Court contracted with Learned Hand in August 2025 after a pilot in which staff benchmarked its output against court work from past cases.
  • The Superior Court of Los Angeles County, the nation's largest trial court, began a pilot in February 2026 with six civil judges and research attorneys under a contract of about $314,000, according to CalMatters. The pilot covers motions for summary judgment and class-action settlement approvals and runs into early 2027.
  • Riverside County Superior Court signed a $10,000 agreement giving seven civil and probate attorneys access.
  • Trial courts in 10 states were also using it as of LawNext's earlier reporting.

Learned Hand becomes "the foundation of its products for courts" inside a judiciary unit Clio has been staffing for weeks: Casetext co-founder Pablo Arredondo joined as SVP, Judiciary in August, and former Michigan Supreme Court Chief Justice Bridget Mary McCormack joined as general manager in September, per the release. That last hire matters for anyone in Michigan: the former chief justice of a court that is a Learned Hand customer now runs the commercial strategy of the vendor.

Why Is One Vendor on Both Sides of the Bench a Problem?

Because Clio's business is selling software and AI to the lawyers whose filings Learned Hand reads. When Clio closed its $1 billion vLex acquisition in November 2025, it reported $400 million in annual recurring revenue and 400,000 legal professionals as customers, and picked up vLex's Vincent AI research assistant. In August 2026 it put Clio Docket — more than 1 billion federal and state court filings — inside Clio Work, so firms can monitor dockets and pull filings straight into their AI workspace.

Now line those up. A litigator drafts a summary-judgment motion in Clio Work. A Los Angeles judge's research attorney summarizes it in Learned Hand, which is designed to flag where the lawyer stretched the record. Both tools belong to the same company, and the release says the plan is to combine them: Learned Hand plus Clio's legal and court data create "the foundation for judicial technology grounded in authoritative law and informed by the record before the court."

Three specific exposures follow:

  1. Data flowing the wrong way. A court's non-public material — sealed filings, draft tentatives, a judge's own writing samples that the tool uses to match style — now sits with a company whose other customers would pay a great deal to understand how a given judge reasons. Nothing in the announcement says that data is walled off from Clio's law-firm products or excluded from training.
  2. Corpus provenance. Learned Hand sold Michigan on being "a closed universe" that draws only from verified legal authorities. If that universe is replaced by vLex and Docket content — the same corpus that powers the advocates' tools — the court should know, and should get to test it again.
  3. Model provenance. CalMatters reported that the tool draws on models from Anthropic, OpenAI and Google. A new owner with its own model stack and its own margin pressure is precisely when underlying models change without a customer notice. We covered how this has already happened on the law-firm side in CoCounsel's New Model Runs on Qwen.

What Is the Strongest Case That This Is Fine?

The strongest case is that legal publishers have always sold to both bench and bar. Courts research on Westlaw and Lexis; so do the lawyers arguing before them. CalMatters' public-records review found roughly a dozen California courts using AI tools from LexisNexis, Thomson Reuters and Microsoft Copilot. Nobody treats that as a conflict. And Clio's argument for the deal is a real one: Jack Newton says that if lawyers move dramatically faster with AI while judges stay at the same capacity, "the bottleneck simply shifts to the courts." Courts genuinely need help, and an owner with Clio's balance sheet may keep the product alive longer than a startup would.

That argument holds for a research database. It breaks for this product. A citator returns public law to whoever asks. Learned Hand ingests the court's own record, learns a specific judge's prose, and writes the first draft of the order. LA County's own court CEO has acknowledged that state rules only require judges to consider disclosing generative AI use, and both LA and Riverside refused to confirm whether litigants know the tool is being tested on their cases. (LA's spokesperson told CalMatters that testing is done on motions already decided, separate from live cases. That protects today's litigants. It does nothing for the court data the tool already holds, or for the expansion the contract maps out.) When the litigant cannot see the tool, the contract is the only safeguard left.


Do California's Court AI Rules Already Cover This?

Only partly, and the gap sits exactly where Learned Hand operates. California's Rule of Court 10.430, effective September 1, 2025, required courts that permit generative AI to adopt use policies by December 15, 2025, and bars entering confidential or nonpublic information into a "public generative AI system" — defined as one that lets anyone other than court staff or judicial officers access the data the court inputs. Two problems:

  • The rule's policy requirement excludes judicial officers' "adjudicative role," which is handled by a softer standard. Drafting a ruling is the adjudicative role.
  • Whether a vendor's system counts as "public" turns on who can access court data. That is a contract term. If the acquirer's terms of service, subprocessor list or training policy change, a system that qualified as non-public last quarter may not this quarter — and nobody at the court will be notified unless the contract requires it.

The failure mode is not hypothetical. In October 2025, two federal district judges in Mississippi and New Jersey told the Senate Judiciary Committee that staff use of generative AI had produced error-ridden orders. Purpose-built tools with source links are an answer to that. A change of owner is the moment to confirm the answer still holds.

What Should Court IT and Law Firm Counsel Do Now?

Treat this as a change of control on a system that touches the adjudicative record, and paper the gaps before anything renews or expands. The LA contract includes a roadmap to test the tool in criminal, family and probate divisions — the most sensitive filings a court holds. That expansion should not happen on a pledge.

This Week:

  1. Pull the contract and the data-processing terms for every Learned Hand deployment or pilot, and find the assignment and change-of-control clauses. Note whether the agreement survives assignment to Clio without your consent.
  2. Send the vendor a written question list: which entity now hosts court data, which subprocessors (including model providers) touch it, and whether any of it is accessible to Clio's law-firm products or used to train any model.
  3. Freeze scope expansion. No new divisions, no criminal or family matters, until the answers are in writing.

This Month:

  1. Negotiate a segregation addendum: court data logically and contractually separated from Clio Work, Vincent and Docket; no training or fine-tuning on court inputs or judges' writing samples; no aggregate "judicial analytics" derived from court usage.
  2. Require model and corpus disclosure with notice: a named list of underlying models and authority sources, and 60 days' notice before either changes — long enough to re-run the benchmarking Michigan did before it signed.
  3. Re-check your Rule 10.430 posture (or your state's equivalent) against the new owner's terms, and document why the system still qualifies as non-public.

Before Renewal or Pilot Conversion:

  1. Get exit rights: export of all work product and prompts in a usable format, certified deletion of court data, and a transition period long enough to stand up an alternative.
  2. Decide on litigant disclosure before a lawyer asks for it in a motion. A court that adopts a disclosure policy voluntarily controls the wording.
  3. For law firm general counsel: if your firm uses Clio and litigates in these courts, ask Clio in writing whether any judicial-side data or insight can reach your workspace. You want the answer to be no, and you want it on file before opposing counsel raises it.

If you are working through similar terms on other acquired vendors, our pieces on vendor neutrality after Visa bought BioCatch and Mark43's one-sentence cross-vendor pledge walk through the same clause-by-clause approach.

The Bottom Line

Legal tech is rebuilding the vertical integration of the old publishing era, with one difference: the product no longer just retrieves the law, it drafts the ruling. Clio can be a careful steward of Learned Hand. It may build exactly the walls courts need. But "continue supporting existing deployments and pilots" is a sentence about uptime, not independence, and independence is the one thing a court cannot outsource.

Judges are required to be impartial. Their software vendor is not. Write the difference into the contract.

Continue Reading

Share:

Frequently Asked Questions

What is Learned Hand?

Learned Hand is an AI workspace for judges and court clerks. It connects to court case management systems and helps summarize filings, research disputed issues, and draft bench memos and orders. Clio acquired it on September 30, 2026.

Which courts use Learned Hand?

The Michigan Supreme Court contracted for it in August 2025, the Superior Court of Los Angeles County began a roughly $314,000 pilot with six civil judges in February 2026, Riverside County signed a $10,000 agreement, and trial courts in 10 states had also used it.

Did Clio commit to keeping court data separate from its law-firm products?

No. Clio's announcement says it will continue supporting Learned Hand's existing deployments and pilots, but it says nothing about segregating court data from Clio Work, Vincent or Clio Docket, about training on court inputs, or about exit rights.

What should a court using Learned Hand ask for after the Clio acquisition?

A written data-segregation addendum, a ban on training with court inputs and judges' writing samples, a named list of underlying models and authority sources with notice before changes, and exit rights covering export and certified deletion — before any renewal or scope expansion.

Newsletter

Stay Ahead of the Curve

Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.

Subscribe

Latest Articles

View All →